On Sat, Aug 17, 2013 at 06:22:55PM -0700, ar16 wrote:
On Sat, Aug 17, 2013, at 04:15 PM, Cristian Morales Vega wrote:
I guess the TCP options being under "ipv4" is just a historical accident. The same options will continue having the same effect even if you use IPv6.
My understading was that the IPv4 & IPv6 stacks are completely separate -- code & function.
Given the long list of separate, IPv6-specific, other-than-tcp sysctls, shared tcp tuning parameters seems odd.
Most of the network layer code is differrent in IPv4 and IPv6. Most of the code implementing the TCP transport layer is shared between IPv4 and IPv6, so the same settings apply I don't think it's odd to have these parameters shared. What's odd is that they are located under net/ipv4 -- I like the term "historical accident" ;)
http://www.spinics.net/lists/netdev/msg246716.html
from Jiri Bohac @ SUSE Labs, Suse CZ, which proposes to introduce separate sysctls for tcp_syncookies_mss_table in *separate* net.ipv4 & net.ipv6 namespaces.
Exception to the above, there is separate code for IPv4 tcp
syncookies and IPv6 TCP syncookies. It felt really strange
putting ipv6-specific settings under net/ipv4, this felt less
evil. But having a single TCP setting under net/ipv6 is strange
as well.
The patch got litle attention so far, so I have yet to see what
people say to this detail.
Regards,
--
Jiri Bohac